3
$\begingroup$

Suppose that $\omega$ and $\eta$ are two distinct primitive $n$-th roots of unity, where $n > 2 \in \mathbb{N}$ and $\omega \neq -\overline{\eta}$. I'm wondering how to prove, using only algebraic-like methods, that $\omega - \eta \notin \mathbb{Q}$. I think that an analysis work would do the job but I want to use Galois-theory properties and field theory to do this question. Can you suggest me how to do it? Also hints would be gladly accepted.

$\endgroup$
  • 1
    $\begingroup$ You need more conditions, as for example $\;1,\,-1\;$ are distinct $\;n\,$-th roots of unit for any even $\;n\;$ , yet $\;1-(-1)=2\in\Bbb Q\;$ ... $\endgroup$ – DonAntonio May 21 '18 at 15:54
  • $\begingroup$ True thing, I added $n>2$ and the fact that roots are primitive. $\endgroup$ – Alberto Andrenucci May 21 '18 at 15:55
  • $\begingroup$ I think if you let $\eta = -\overline{\omega}$ you still have a counterexample. $\endgroup$ – B. Goddard May 21 '18 at 15:56
  • 2
    $\begingroup$ It seems artificial not to use your proof "clearly an analysis work would do the job" - how would you prove it? $\endgroup$ – Dietrich Burde May 21 '18 at 16:03
  • $\begingroup$ Sure thing, I should have said that I think that an analysis work would do the job (we can analitically write down in terms of sines and cosines so i think we can actually do the computation). I didn't do the problem in that way but I'll think about it. $\endgroup$ – Alberto Andrenucci May 21 '18 at 16:12
4
$\begingroup$

Here's an entirely algebraic proof. However, I do include a variant using basic properties of $\Bbb C$ to derive equation $(**)$ because that's how I originally came up with it.

Wlog, your two primitive $n$-th roots of unity are $\omega$ and $\omega^r$ with $2\le r\le n-1$ and $gcd(r,n)=1$. Galois theory of cyclotomic fields tells us the Galois group of $\Bbb Q(\omega)|\Bbb Q$ consists of the automorphisms induced by $\omega \mapsto \omega^s$ for $gcd(s,n)=1$. So $$\omega^r-\omega \in \Bbb Q$$ is equivalent to $$\omega^{sr}-\omega^s = \omega^r-\omega \text{ for all } s \in \Bbb Z \text{ with } gcd(s,n)=1. \quad (*)$$

Now in particular we can choose $s=-1$ and get

$$\dfrac{\omega^{-r}-\omega^{-1}}{\omega^r-\omega} =1 ;$$

multiplying both sides with $\omega^{r+1}$ gives

$$-1 = \omega^{r+1} \quad(**).$$

Thus $n | 2(r+1)$. Now $n=r+1$ would contradict $(**)$, so $n=2r+2$, which is even, so $r$ is odd, so $$4|n \text{ and } r=\frac{1}{2}n-1 \quad (***).$$

Now setting in particular $s=r$ in $(*)$ and noting that, by $(***)$, $r^2 \equiv 1$ mod $n$, gives $\omega-\omega^r = \omega^r-\omega$, thus $\omega = \omega^r$, contradiction.

(Note also that your extra condition $n>2$ is vacuously true / redundant since you demand $\omega \neq \eta$. The condition $\omega \neq - \bar\eta$ is a bit weird because actually, $\omega = - \bar\eta$ is the only interesting case, namely it's equivalent to $(**)$. Actually, $\omega \neq - \bar\eta$ is equivalent to the much weaker $\omega -\eta \notin \Bbb R$ -- or in other words, $(**)$ is equivalent to $\omega = -\bar\eta$ is equivalent to $\omega -\eta \in \Bbb R$. See next paragraph.)


(Proof of $(**)$ with complex numbers:) Choosing any embedding of those unit roots into $\Bbb C$, their difference being $\in \Bbb Q$ means it's $\in \Bbb R$, hence their imaginary parts are equal. Because they both lie on the unit circle, this implies $$-1 = \omega \cdot \omega^r = \omega^{r+1} \quad(**)$$

The above proof avoids this, but of course the automorphism $\omega \mapsto \omega^{-1}$ "is" complex conjugation.


Finally note that if only $\omega$ is demanded to be primitive, one has the counterexample $\omega^2-\omega=-1$ for $n=6$. (Which brings up a different proof idea: If $\omega^r-\omega = a \in \Bbb Q\setminus \{ 0\}$, then the $n$-th cyclotomic polynomial has to divide $X^r-X-a$, which one can maybe lead to a contradiction too.)

$\endgroup$
  • $\begingroup$ Nice thoughts and nice answer! Thanks a lot! $\endgroup$ – Alberto Andrenucci May 23 '18 at 13:52
3
$\begingroup$

That's true for any two roots of an irreducible polynomial $P(x)\in \mathbb{Q}[x]$. If we had $x_1+r=x_2$, for $r \in \mathbb{Q}$, $r\ne 0$, then the polynomials $P(x)$ and $P(x+r)$ would have a common root $x_1$, so, being irreducible, must be the same, contradiction.

$\endgroup$
  • $\begingroup$ That's a neat, strong statement and proof (+1). Still I'm happy I went through the special case with special arguments, if only because that gave the extra info about when the difference is in $\Bbb R$. $\endgroup$ – Torsten Schoeneberg May 24 '18 at 1:47
  • 1
    $\begingroup$ Thank you! Yes, that is an interesting question. Btw, the quote from Pascal is nice. $\endgroup$ – Orest Bucicovschi May 24 '18 at 2:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.